Design And Implementation Of An Optical Detection System For Apple Surface Damage

Apple is the main agricultural product in our country and one of the fruits that people often eaten in daily life.At present,the sale value of apple has been affected by the easily bruise on the surface during picking,sorting,and transportation.Therefore,the research and identification of apple bruise had an important effect on quality grading of apples.The traditional method of identifying apple bruise not only requires a lot of human labor,but increases production costs.And the effect of manual sorting could be worse because of the long-term repetitive labor,which caused by visual fatigue.Thus,how to identify bruise on the surface of apples quickly and accurately has become an important issue facing on researchers.Based on hyperspectral technology,a set of bruise identification optical detection online system was designed in this research,which provides a new solution for Apple's non-destructive testing.The main contents of this paper are as follows:(1)The based on the analysis of several commonly used optical detection principles and advantages and disadvantages,55 apple hyperspectral images in three time periods(0,12 and 24 hours)were collected by the hyperspectral imaging system,which had bruise on surface,and the spectral data were processed.After the minimum noise separation transform is used to reduce the dimensionality of the apple sample data,the Otsu algorithm,average pixel method,image segmentation,morphology,and maximum connected area are used to determine the region of interest of the apple,and the region of interest is determined by the resampling method.Divide into 8 equal parts and obtain the average spectrum of each equal part.Five preprocessing methods are used to correct the average spectrum,and four classification algorithms are combined to establish a prediction models.The research showed that the LightGBM prediction model had the best classification performance after combined with multiplicative scattering correction.The bruise recognition rate under the full spectrum data reached 96.41%,and the bruise recognition rates of the three time periods reached 97.08%,96.11% and 95.81%.(2)Using the successive projections algorithm to extract the characteristic wavelengths of the spectral data.According to the importance of the characteristic selected 11 characteristic wavelengths and established the prediction model based on the relationship between the number of selected characteristic wavelengths and the accuracy of the model.The research showed that the overall recognition rate of the LightGBM model reached 95.79%,and the bruise recognition rate for 0h,12 h and 24 h reached96.63%,95.97% and 94.77%.(3)The LightGBM model was selected to identify the bruise of apples,and based on this,an online detection system was designed.The system was mainly divided into two parts: the first part was the client,which main functions were to import the sample data set,link the sever remotely and display the prediction results;the second part was the server,which was built on the cloud server and mainly completed the function is predict the result of sample classification and sending the prediction results to the client using TCP/IP communication.The experimental research showed that the online inspection system for apple bruise identification designed in this research could detect the bruise on apple surface effectively,which provided an idea for rapid detection of apple bruise.